rfoo

16 ms·

151.
by rfoo 2y ago
I still believe that larger models are better at covering the long tail. Our benchmarks are saturated, but actual model capability is not.
152.
by rfoo 2y ago
All good questions. > is not modifying those code to use STL containers much cheaper That's right. However, I'd add that most exploited bugs these days (in high-profile targets) are temporal memory safety (i.e. lifetime) bugs.
153.
by rfoo 2y ago
My two cents, I'm wearing my exploit writer's hat, but my current day job is SWE on legacy/"modern-ish" C++ codebases. > Enable STL bounds checking using appropriate flags This rarely helps. Most of the nice-to-e
154.
by rfoo 2y ago
Besides functional equivalence, a significant part of the value in neural decompilation is the symbol (function names, variable names, struct definition including member names) it recovered. So, if the LLM predicted "FindFirstFitContai
155.
by rfoo 2y ago
I've been thinking on how to build a benchmark for this stuff for a while, and don't have a good idea other than LLM-as-judge (which quickly gets messy). I guess there's a reason why current neural decompilation attempts are
156.
by rfoo 2y ago
tbh the "short the stock market" story is pretty silly, it wasn't predictable at all. but yeah, the guy got to do whatever he want to do now.
157.
by rfoo 2y ago
Now that Noam is back I'm a little bit more optimistic.
158.
by rfoo 2y ago
ByteDance has been working on autoregressive image generation for a while (see VAR, NeurIPS 2024 best paper). Traditionally they weren't in the open-source gang though.
159.
by rfoo 2y ago
Ha, I still remember that super hilarious "You are under 18, so you should not write C++, as it is unsafe..." log from ... a year ago?
160.
by rfoo 2y ago
> Right now DeepSeek models hosted in china are having very high latency. If you are talking about DeepSeek's own hosted API service. It's because they deliberately decided to run the service in heavily overloaded conditions an
161.
by rfoo 2y ago
In what respect is US one single large country? Texas and California feels as different as Shandong and Shanghai. And not sure if a flame war will happen if I try to compare Hawaii and Taiwan :p
162.
by rfoo 2y ago
Because they want to advertise that they CAN break into phones loaded with GrapheneOS. Just not the latest version. Likewise they "announce" that they can't unlock BFU iPhone-s running latest iOS, and the real message is &quo
163.
by rfoo 2y ago
> that every disk will experience a fatal failure or a disconnection at least once a month When? I vaguely remember that it used to be like that, but I haven't seen nearly as many failures on Aliyun for the last few years.
164.
by rfoo 2y ago
There is CONFIG_STATIC_USERMODEHELPER that disables the sysctl you mentioned and actually make modprobe_path read-only.
165.
by rfoo 2y ago
If anything, it sounds more like an "EA working as intended" story.
166.
by rfoo 2y ago
> why would they be running such an old Linux? They didn't. OP misunderstood what gVisor is, and thought gVisor's uname() return [1] was from the actual kernel. It's not. That's the whole point of gVisor. You don&#x27
167.
by rfoo 2y ago
tl;dr whichever system using an ESP32 as a bluetooth adapter may also just run arbitrary code on the ESP32 itself over the same interface. Commands have to be issued from the host system, not from the air. This sounds like... a good feature
168.
by rfoo 2y ago
For decode, MoE is nice for either bs=1 (decoding for a single user), or bs=<very large> (do EP to efficiently serve a large amount of users). Anything in between suffers.
169.
by rfoo 2y ago
> Still looks compute-bound to me. H100 has 3.3TB/s HBM bandwidth on paper, and ~1000TFLOPS bf16 compute on paper. That's 1:300. 0.6GB vs ~2GFLOPS is 1:3. Tell me how is this compute bound? (also, your number, even after accoun
170.
by rfoo 2y ago
> the vast majority of speedrunners are "only" practicing and replicating exploits demonstrated by others So they are red teamers :p
171.
by rfoo 2y ago
Yeah, I don't want to do this either. This is a super special case, after exploring alternatives with our researchers it's unfortunately needed. As for record-of-death, we made sure that we do serialize all rng state and have our
172.
by rfoo 2y ago
They did this back in their trading firm days, and... Imagine that you have a sequence of numbers. You want to randomly select a window of, say, 1024 consecutive numbers, a sequence, as input to your model. Now, say, you have n items in thi
173.
by rfoo 2y ago
Ah, makes sense. Sadly RDMA isn't that fast for now, or at least commercial RNICs/switches don't :( Once you left your host in data center network, everything counts in microseconds.
174.
by rfoo 2y ago
> What makes the workload somewhat special is I'll add that latency also doesn't matter that much. You are doing batched data loading for batch n+1 on CPU when GPUs are churning batch n-1 and copying batch n from host memory at
175.
by rfoo 2y ago
That's been illegal since May 1995 (before that China had six working days week). Does it really matter whether it's illegal or not, if there is no enforcement? Pinduoduo (in other name, Temu) has been doing 70 hours week since th
176.
by rfoo 2y ago
Huh? What kind of RDMA has a completion latency of 20 nanoseconds? It's more like 5 microseconds. I agree that a lot of "modern" storage stack is way too slow though, tried to find a replication-first object storage for crazy
177.
by rfoo 2y ago
I don't think so. This looks like to be using an actual VM instead of container tech.
178.
by rfoo 2y ago
Check out why Togerther.AI acquired CodeSandbox.
179.
by rfoo 2y ago
I agree. Gamers are cursing Nvidia right now tho, and sadly university labs doing serious research on gaming cards is also a past :(
180.
by rfoo 2y ago
I think we did release some of the optimized kernels but I don't think we have released any one with SASS black magic, at least not before I left. Already been sanctioned by BIS, better not annoy NVIDIA furthermore.

More ›